Proceedings of the Sixth International Conference on Machine Learning and Cybernetics, Hong Kong, 19-22 August 2007
A NOVEL ADMINISTRATIVE MODEL FOR COLLABORATIVE LEARNING
SYSTEM
RONG ZHANG, LI-JUN GUO, ZHEN LIU
Faculty of Information Science & Technology, NingBo University
E-MAIL: {zhangrong ,guolijun ,liuzhen}@[Link]
Abstract: problem of interaction between learners in web-based
In a collaborative learning setting, learners and tutors collaborative learning environment to some extent. There
need some collaborative and controlling means to harmonize are many practical questions, however, which need new
their learning actions. For this purpose, a novel administrative solutions for technological implementation. In fact, the
model for collaborative learning environment is presented in exiting problems, among which are: the problem of disorder
this paper, which constructs shared task space by
decomposing learning task and reasonably allocating learning
in hypertext information, the problem of users’
resource. This paper analyses the elements of task space, disorientation, and the problem of being unable to
presents a set of formal definitions of task space model, and harmonize and control the learning processes of learners
introduces the model’s implementation. Finally, the paper when fulfilling a certain learning task, can be possibly
summarizes the model’s characteristics in the collaborative solved by constructing effective management model for
learning system and highlights the practical significance of the collaborative learning.
model. Generally, a good collaborative learning environment
should be able to well integrate learning materials with
Keywords: leaning tools and leaning task, and to effectively solve the
CSCL; Management model; Task space; Learning issue of misguiding the learners [4-5]. Additionally, it can
category provide imperative awareness information to facilitate the
learners to communicate with each other as well as
1. Introduction improved knowledge sharing. In the light of the
characteristics of collaborative learning system, we present
Computer supported collaborative learning (CSCL) a novel collaborative learning administration model based
has grown as a result of integrating the theories and on dynamic task space. In this paper, the formal definitions
methods of computer supported collaborative work (CSCW) of task space model are presented first. Approaches to
with that of collaborative learning (CL). Since its presence construct the task space are given in detail in section 3,
and implementation, CSCL has been based on the promise followed by the design and implementation of the
of fundamentally changing the traditional ways in which management tools corresponding to this model. At last, this
people have been learning [1]. Supported by CSCL, paper is concluded by characteristics of the model, a
teachers and students can overcome the limit of time and research project on it being carried out at Ningbo
space to carry out their collaborative learning activities University.
such as synchronous distance learning, asynchronous
collaborative learning and group discussions, which can 2. Basic Conceptions
arouse students' initiative and increase learning efficiency.
At present, main research work on CSCL include support By further decomposing a learning task for a specific
system for CSCL, environment management model of goal into subtasks, the task-space-model specifies the
CSCL, analysis and assess on collaboration in collaborative fundamental elements of the task and the relevant rules
learning, etc [2-3]. between the sub-tasks, and simultaneously provides
As for the research on computer-supported learners with controllable learning processes (learning
collaborative learning system, a number of CSCL systems activities) under the corresponding learning model.
have been constructed, such as Web CT at University of Moreover, multi-level awareness space is available by
British Columbia, Habenero at University of Illinois, and constructing subspaces, which facilitates learners to
Virtual-U, web-based learning tools, at University of Simon communicate with each other during a learning process.
Fraser in Canada [3-4]. These systems have solved the
1-4244-0973-X/07/$25.00 ©2007 IEEE
4159
Authorized licensed use limited to: JADAVPUR UNIVERSITY. Downloaded on March 06,2010 at 03:58:26 EST from IEEE Xplore. Restrictions apply.
Proceedings of the Sixth International Conference on Machine Learning and Cybernetics, Hong Kong, 19-22 August 2007
Following are the definitions of the elements for the task relations of all subcategories, D is the available learning
space: resources of the space and is described as a information
Definition 1: A learning category is a relatively subspace[6] which will be further explained later.
independent learning goal in a given range. A specific
learning category may be involved into a subject, a 3. How to Establish Dynamic Task-Space Model
speciality, a course, a learning cell or a knowledge point, all
of which are on five different levels respectively. There are According to the definitions above, this section will
three kinds of mutuality, i.e. overlay, sequence and parallel, describe how to establish a model on three levels: category,
between the learning categories on different levels and even task (activity) and task space. The process of modeling for a
on the same. There are four states for a learner in a learning dynamic task space illuminates the process of collaborative
category, i.e. “incapable”, “not learned”, “being learned” learning under the model's administration.
and “passed”.
Definition 2: Overlay relation between learning 3.1. Specifying Learning Categories
categories refers to the direct affiliation relation between Ultimately, a learning category is specified
two learning categories, generally which are on two independent of task, beginning with higher levels. A
different levels. For instance, Operating System (OS, course) specified learning category includes the following
and memory management (learning cell) are two kinds of information: unique identifier, level, description of learning
learning category that belong to the relation of overlay. The goal, affiliation, state, etc, where level and description of
overlay relation between learning categories is static and learning goal can significantly help establish learning tasks
does not varies depending on the learning process. dynamically, affiliation of the learning category is the
Definition 3: Sequence relation between learning criterion to set up the activity diagram of learning tasks, and
categories refers to the relation that the learning category X state of the learning category is decided dynamically
must be learned before the learning category Y, generally according to learning patterns of the tasks and changes in
which are on the same level. For instance, in the course of activity chain during a learning process
OS, process management (learning cell) and process
communication (learning cell) would be deemed to belong 3.2. Specifying Tasks
to this relationship since process management must be After specifying learning categories, learning tasks are
learned prior to process communication. Generally, the specified and an elementary activity diagram corresponding
sequence relation between learning categories is static, but to the tasks is constructed basing on correlativity between
occasionally it can vary according to learning task. the categories. An activity diagram is directed, in which
Definition 4: Parallel relation between learning nodes represent learning categories on different levels. In
categories refers to the relation that during a learning an activity diagram, overlay relation between learning
activity, the sequence to be learned of two learning categories is represented by a sub-diagram, sequence
categories on the same level are not fixed. For instance, relation is represented as precedent node and successor
during the learning process of the category of OS, storage node, and parallel relation is represented as nodes on the
management (learning cell) and I/O system management same level which are not precedent and successor for each
(learning cell) can be learned in arbitrary order. The parallel other.
relation between learning categories is established In the case of the learning categories of OS, the
dynamically and varies with learning task. process of building an activity diagram for it is described
Definition 5: A task involves a series of learning below:
activities aiming to accomplish the desired learning goal for 1. Specify learning categories, list the categories on
a specific learning category, necessarily and reasonably the immediate lower level, and build the initial activity
decomposing the learning category into different levels diagram (see Figure 1);
(task decomposition) and adopting corresponding
collaborative learning pattern for the sub-categories on …
different levels. Ti
T1
Definition 6:A task space is constructed for a specific
learning category, characterized by a quadruple: S T0
(T)={ Tid, Cs, N, D}, where given the task space S (T), Tid … Tj
is the identifiers of the learning category represented by the T2
task space, Cs is a set of categories related to the given Figure 1. Initial activity diagram
category, N represents the network frame reflecting the
4160
Authorized licensed use limited to: JADAVPUR UNIVERSITY. Downloaded on March 06,2010 at 03:58:26 EST from IEEE Xplore. Restrictions apply.
Proceedings of the Sixth International Conference on Machine Learning and Cybernetics, Hong Kong, 19-22 August 2007
2. Select correlative learning categories on the same 3.3. Establishing Dynamic Task Space
level in terms of tasks' goals, roles of participation and
learning patterns (teaching pattern, discussion pattern and Upon that, content of three tuples in the quadruple
self-studying pattern), etc., and establish corresponding (Tid , Cs , N , D)constituting task space, which are Tid
relations of the nodes in the activity diagram built in 1 (see (category corresponding to task), Cs (a set of correlative
Figure 2); categories) and N (correlativity framework of categories),
have been defined. The last tuple undefined is information
… subspace including learning resources and interactive tools.
Ti The intentions of establishing dynamic task space basing on
T1 specified tasks are to establish corresponding information
T0 subspace and to allocate learning resources and tools to
… Tj nodes in directed activity diagram, and simultaneously, to
T2 define essential constraint relationship so as to restrict and
describe activities of learning categories. Constraint
Figure 2. Activity diagram with defined basic relations relationship is described by states of learning schemas and
3. Build sub-diagram of activity for each node level by learning categories.
level in terms of goals of tasks (see Figure 3). Sub-diagram
of activity is also constructed by performing step 1 and step 4. Design and Implementation of Administrative
2 above. In our prototype, constructing the activity diagram Tools of the Model
is fully completed by administration tools of the model.
It is administrative tools of the model that play a key
role in successfully applying dynamic task space model to
collaborative learning and meanwhile overcoming the
deficiencies of previous models for collaborative leaning.
T11 T12 …
T1 In this section, we describe data structure, operation
Ti algorithms and implementation of administrative tools of
T0 T22 … the model.
Tj
T21 T24
T23 T2 4.1. Data Structure in the Model
In the dynamic task space model, the main line is task
Figure 3. Activity diagram with sub-diagram and the main body of task is directed activity diagram.
Different learning categories are connected and nested into
T: root node of activity diagram，represents the course of
OS
directed activity diagram according to correlativity.
T0: first node to root node ， represents fundamental
Therefore, a task space model can be described by the
concepts of OS following data structure:
T1: successor node directly to T0，represents processor struct TaskNode{
management int Tid; string Description;
T11: first son-node of T1，represents process model TaskNodeLink *preNode,*nextNode;
T12: son-node of T1，represents thread model TaskNode* SonNode;
T2: successor node directly to T0，represents memory TaskNode* Parent;
management int Method;
T21: first son-node of T2，represents the concept of int infoTid;
memory int state;
T22: son-node of T2，represents successive schema }
T23: son-node of T2，represents non-successive schema Tid is node identifier in activity diagram. Description
T24: son-node of T2，represents instance of memory is description of the category represented by a node.
management preNode and nextNode respectively represents immediate
Ti, Tj: son-node of T，some categories in later period precedent node chain and successor node chain to the
current node. The data type of TaskNodeLink is given
below. Relation of precedent and successor between nodes,
4161
Authorized licensed use limited to: JADAVPUR UNIVERSITY. Downloaded on March 06,2010 at 03:58:26 EST from IEEE Xplore. Restrictions apply.
Proceedings of the Sixth International Conference on Machine Learning and Cybernetics, Hong Kong, 19-22 August 2007
which is the main basis for topological sorting of nodes in else
the same layer, represents the sequence relation between //insert nodes into the sub diagram of iParent
learning categories. Further, SonNode points to the first { while [Link]<>nil
son-node, null value of which indicates that there are no iPreNode:= [Link];
nodes on the next layer, i.e., the corresponding category is //find out the tail in the successor chain of iPreNode
not decomposed; not null value of which indicates that [Link] := t;
there is a nested sub activity diagram, i.e., the //add new node into successor chain of precedent node
corresponding category has been discomposed into some while [Link]<>nil
sub categories on the next layer. Other son nodes are iNextNode:= [Link];
accessed by the successor nodes to the first son node. //find out the tail in precedent chain of iNextNode
Parent points to the parent node. Method and state are [Link] := t;
restrictions on category during fulfilling learning task, //add new node into precedent chain of successor node
specified restriction values of which have been described }
above. At last, infoTid is the identifier of sub information }
space, by which different learning categories are restricted Algorithm 2: Visiting an activity diagram begins with
to access to corresponding learning resource. root node and is to traverse nodes in a depth-first order.
struct TaskNodeLink{
TaskNode *pTN; Void DFSTraverse( TaskNode *pT)
TaskNodeLink *next; { if [Link] == nil
} visit(pT);
This structure given here is used to build up a //visit a node, fulfill learning of the current learning
precedent node chain or a successor node chain. // category and reset the state of it
else{
4.2. Operation Algorithms and Implementation Topologicalorder(pT);
// get a topological order set of son nodes and return x
Basing on the data structure above, we can describe an // with Stcollection
activity diagram of learning category on any level. The For index in Stcollection
algorithms of activity diagram include establishing an DFSTraverse(index);
activity diagram and visiting an activity diagram, which are }
illuminated by Algorithm 1 and Algorithm 2 below //do some summary work before end the learning of
respectively. // the current category
Algorithm 1: In this administration model, each }
activity diagram corresponds to a category on the highest There are two processes called, visit() and
layer, i.e. there is only one root node which has no parent Topologicalorder(), in Algorithm 2. In visit(), learning of a
node and represents the leaning category on the highest learning category can be accomplished by the attributes of
layer. When a user inserts a new node into activity diagram infoSID and Method of nodes as well as descriptions of
using interface of the administration model, he need specify learning goals, and furthermore, a learning task can be fully
parent, precedent and successor to this node. Activity accomplished by setting states of learning categories and
diagram is built up by function Creat(). repeatedly visiting activity diagrams of learning categories.
Create(String Description,TaskNode* iParent, *iPre In addition, basing on the storage structure of activity
Node, *iNextNode) diagram, it is easy to get topological order of all son nodes
{ in terms of SonNode 、preNode and nextNode. This job is
TaskNode* t=new TaskNode (NextID(), Description, accomplished by Topologicalorder().
iPreNode,iNextNode,iParent,0,0);
if iParent == nil //create root node in activity diagram 4.3. Implementation of Administration Tools
return
else The collaborative learning system based on this model
if [Link] == nil consists of four main parts as illustrated in Figure 4.
//create sub diagram of iParent and add first node into Effective administration tools of dynamic task space model
//the sub diagram are the foundation and precondition to develop the system.
[Link] := t Considering operational feasibility, these administration
4162
Authorized licensed use limited to: JADAVPUR UNIVERSITY. Downloaded on March 06,2010 at 03:58:26 EST from IEEE Xplore. Restrictions apply.
Proceedings of the Sixth International Conference on Machine Learning and Cybernetics, Hong Kong, 19-22 August 2007
tools are implemented in interactive mode with graphic supports knowledge sharing by recording learners’
collaborative learning course; 4) Flexibility and dynamic
feature of task space increase concretion, pertinency and
operatability of collaborative learning under this model. We
have applied this model and administrative tools to
implementation of a collaborative learning system as the
outcome of a foundation project at Ningbo University and
good results have been attained in our research.
Figure 4. Implementation of the model
Acknowledgements
interfaces. While professional teachers creating all learning
categories in a knowledge area, administration tools with I'd like to sincerely thank all the colleagues involved in
graphic interface provide relevant operation, such as the research project, who are Prof. Guo, Dr. Liu as well as
creating son node, setting precedent node and setting the graduate students.
successor node, to establish activity diagrams. In this model,
the dynamic feature of task space is embodied in two References
aspects: First, dynamically establishing learning tasks in a
learning category, .i.e. establishing the dynamic activity [1] Chen Q, Liu R, Zhang J, “Creating a constructive
diagram, help teachers define different leaning sequence of environment for learners to explore word problem
learning categories on different levels according to his(or solving”, Proceeding of International Conference on
her) own apprehension; Second, in sub activity diagram, Computers in Education, Beijing, Vol.2, pp. 498-503.,
dynamic topological order of learning categories on the 1998.
same level, which is defined during learning, makes every [2] Cerardo A, “A collaborative Learning environment
learning group follow different learning steps as long as based on intelligent agents”, Expert Systems with
there is no violating correlativity between learning Applications, 14(1-2), pp. 129-137, 1998.
categories. Moreover, administration tools can record [3] Bruce M e, Chan L K S. “Co-operative learning in
learning history of individual learning group, which helps a integrated classroom”, Curriculum and Teaching, 6(1),
group learn repeatedly and learn with other groups by pp.48-52, 1991.
sharing information and experience. [4] Celine B, “Complementarity of information and
quality of relationship in cooperative learning”, Social
5. Conclusions Psychology of Education, 4(3/4), pp.335-357, 2001.
[5] Benford S D , Fahlen L E, “A Spatial Model of
In the dynamic task space model, elements of learning Interaction in Large Virtual Environments”,
task and related rules of tasks are specified by task Proceeding of Third European Conference on
decomposition of specified learning goal. A collaborative CSCW(ECSCW93), Milano, Italy, Kluwer, pp.
learning environment based on this model has the following 109-124.
[6] Wand Jinfeng, Zhan Yongzhao, Mao Qirong, “Nested
strong points: 1) It can effectively resolve the problem of
learners' disorientation; 2) It facilitates information Knowledge Space Model and Implementation for the
communication among learners by sub information space to Collaborative Learning Environment”, Computer
Science,31(5), pp.150-152, 2004.
provide learners with multi-layered awareness space; 3) It
4163
Authorized licensed use limited to: JADAVPUR UNIVERSITY. Downloaded on March 06,2010 at 03:58:26 EST from IEEE Xplore. Restrictions apply.